An H-bridge drives a DC motor in both directions with four switches, usually MOSFETs. At any moment two of them are on, one high-side and one low-side, and current flows through both. Every bit of their resistance turns into heat, and heat is what kills motor drivers.
Conduction loss
A MOSFET that's fully on behaves like a small resistor, R_DS(on). Its loss is I²R, and with two in the current path:
P = 2 × I² × R_DS(on)
For a 10 A motor through MOSFETs rated at 25 mΩ:
P = 2 × 10² × 0.025 = 5 W total, 2.5 W in each conducting MOSFET
| Motor current | R_DS(on) | Total loss | Per MOSFET |
|---|---|---|---|
| 2 A | 25 mΩ | 0.2 W | 0.1 W |
| 5 A | 25 mΩ | 1.25 W | 0.63 W |
| 10 A | 25 mΩ | 5 W | 2.5 W |
| 10 A | 10 mΩ | 2 W | 1 W |
| 20 A | 10 mΩ | 8 W | 4 W |
| 10 A | 100 mΩ | 20 W | 10 W |
Because of the square, doubling the current quadruples the heat. And R_DS(on) matters as much as current: the 100 mΩ part makes ten times the heat of the 10 mΩ one.
R_DS(on) rises when the MOSFET is hot
The datasheet's headline R_DS(on) is at 25 °C. At a junction temperature of 100 °C it's typically about 1.5× higher (check the "normalised on-resistance vs temperature" graph). So the 5 W example is really closer to 7.5 W once it warms up. And because more heat means more resistance, a marginal design can run away thermally.
Also check the gate voltage. R_DS(on) is quoted at a specific V_GS, often 10 V. Drive a standard MOSFET from a 3.3 V microcontroller pin and it may barely turn on. Use a logic-level part rated at 4.5 V or below, or a gate driver.
Switching loss
Every time a MOSFET turns on or off with PWM, it spends a few tens of nanoseconds with both voltage across it and current through it. A rough estimate per switching MOSFET:
P_sw ≈ ½ × V × I × (t_rise + t_fall) × f_PWM
At 12 V, 10 A, 100 ns total transition time and 20 kHz PWM: ½ × 12 × 10 × 100 ns × 20,000 = 0.12 W. Small next to conduction loss here, but it grows with PWM frequency and supply voltage, and a weak gate drive can make the transitions ten times slower.
Does it need a heatsink?
Temperature rise is power times thermal resistance:
T_junction = T_ambient + P × R_θJA
A D²PAK MOSFET on about one square inch of PCB copper has an R_θJA of roughly 40 °C/W. At 2.5 W that's a 100 °C rise, too hot in a 25 °C room plus the R_DS(on) increase above. Options, roughly in order of cost:
- Lower R_DS(on). A 10 mΩ part cuts the heat to 1 W per MOSFET.
- More copper. Larger pours, both layers, thermal vias under the tab.
- A heatsink or airflow.
- Parallel MOSFETs. Two in parallel halve the resistance and spread the heat.
The L298N problem
The popular L298N module uses bipolar transistors, not MOSFETs, and drops roughly 2 V or more across the bridge. At 2 A that's about 4 W or more wasted, against 0.2 W for a 25 mΩ MOSFET bridge at the same current. It's also why your motor sees noticeably less than the supply voltage. For anything over an amp or so, a MOSFET driver (TB6612FNG, DRV8871, BTS7960 and similar) runs far cooler.
Common mistakes
- Counting one MOSFET instead of two. Current passes through a high-side and a low-side switch.
- Using the 25 °C R_DS(on). Plan with the hot value.
- Using the motor's average current when it stalls or starts often: stall current can be 5–10× the running current, and I² makes that 25–100× the heat.
- Forgetting the traces. The PCB copper carrying 10 A has resistance too; size it with the trace width calculator.
Try it
The H-bridge power calculator gives the conduction loss from current and R_DS(on). To estimate the motor current in the first place, use the robot speed and torque calculator, and size the copper with the PCB trace width calculator or the trace width worked example.